diff --git a/src/lib.rs b/src/lib.rs
index b703ac36c..d5aea846a 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -1342,7 +1342,14 @@ impl Build {
         if !msvc || !is_asm || !is_arm {
             cmd.arg("-c");
         }
-        if self.cuda && self.files.len() > 1 {
+        if self.cuda
+            && self
+                .files
+                .iter()
+                .filter(|file| file.extension() == Some(OsStr::new("cu")))
+                .count()
+                > 1
+        {
             cmd.arg("--device-c");
         }
         if compiler.family == (ToolFamily::Msvc { clang_cl: true }) {
@@ -1999,7 +2006,14 @@ impl Build {
             self.assemble_progressive(dst, chunk)?;
         }
 
-        if self.cuda {
+        if self.cuda
+            && self
+                .files
+                .iter()
+                .filter(|file| file.extension() == Some(OsStr::new("cu")))
+                .count()
+                > 0
+        {
             // Link the device-side code and add it to the target library,
             // so that non-CUDA linker can link the final binary.